Cross-linking agent production control system and method based on Internet of Things

Abstract

The present invention discloses a cross-linking agent production control system and method based on the Internet of Things, relating to the field of chemical production control technology. The system is composed of several functional modules, including: a data acquisition module, which acquires spectral data in a reactor, as well as temperature and pressure data at different locations on the reactor wall; a data processing and control module, which converts spectral data into reaction data of a cross-linking agent intermediate based on a meta-calibration model; an image recognition and pattern analysis algorithm is used to identify the status data of the cross-linking agent intermediate based on the temperature and pressure data at different locations, including local overheating areas and abnormal pressure areas; an execution decision module, which establishes a dynamic model integrating the production characteristics of the cross-linking agent based on the reaction data and status data, and uses historical data and real-time IoT data to train an ML model based on the dynamic model to automatically generate and execute the optimal control strategy under safety constraints.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical production control, and in particular to a cross-linking agent production control system and method based on the Internet of Things. Background Art

[0002] In the rubber industry, cross-linking agents reshape the molecular structure of rubber through vulcanization reactions, giving tires excellent wear resistance and anti-aging properties. As core materials in the modern chemical industry, the performance and quality of cross-linking agents directly determine the application performance of downstream products. In the coatings field, cross-linking agents act as the "invisible pillar" of paint film strength and durability, ensuring that the coating can maintain its integrity in complex environments. As the wave of Industry 4.0 sweeps the world, the chemical industry is accelerating its transformation towards intelligence and greenness. The cross-linking agent production process involves complex physical and chemical reactions. The traditional extensive production model can no longer meet the stringent requirements of modern industry for product consistency, production efficiency, safety and environmental protection. From the precise control of raw material ratios to the dynamic monitoring of reaction processes; from real-time optimization of product quality to the intelligent management of energy consumption and emissions, every link urgently needs to be enabled by more advanced technical means.

[0003] In the cross-linker production process, existing technologies have many problems. Traditional component concentration monitoring relies on offline laboratory testing, which takes several hours or even longer from sampling to obtaining results. It cannot reflect the changes in components in the reactor in real time, resulting in serious delays in adjusting the feed ratio and affecting the stability of product quality. In terms of temperature and pressure monitoring, single-point sensors are usually used, which makes it difficult to fully capture the complex temperature and pressure field distribution in the reactor. Especially in the exothermic reaction process of cross-linker synthesis, potential risks such as local overheating and abnormal pressure cannot be discovered in time, which may cause safety accidents and product quality defects. At the same time, the existing production control system lacks the ability to continuously monitor key parameters of the reaction process, and cannot achieve refined and intelligent control of the production process. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0005] The cross-linking agent production control system based on the Internet of Things includes:

[0006] The data acquisition module acquires the spectral data inside the reactor, as well as the temperature and pressure data at different locations on the reactor wall;

[0007] The data processing and control module converts spectral data into reaction data of the crosslinker intermediate based on the meta-calibration model. Based on the temperature and pressure data at different locations, it uses image recognition and pattern analysis algorithms to identify the status data of the crosslinker intermediate, including local overheating areas and abnormal pressure areas.

[0008] The execution decision module establishes a dynamic model of the production characteristics of the fusion cross-linker based on reaction data and status data. Based on the dynamic model, the ML model is trained using historical data and real-time IoT data to automatically generate and execute the optimal control strategy under safety constraints.

[0009] Furthermore, the spectral data is obtained specifically by quantitatively tracking the changes in the intensity of the characteristic peak, recording the absorbance of the characteristic peak of the target reactant, and measuring the absorbance of the same characteristic peak again at the tth minute of the reaction; by comparing the initial value and the real-time measurement value, the consumption ratio of the reactant is calculated until the reaction is completely completed, and the maximum absorbance of the characteristic peak of the product is measured, that is, the spectral data, to monitor the cross-linking agent reaction process in real time.

[0010] Furthermore, the process of obtaining temperature data and pressure data at different positions on the reactor wall is as follows:

[0011] The temperature and pressure of the reactor wall are measured in a distributed manner through a flexible sensor array on the outer surface of the reactor wall; the measurements are output in the form of digital signals and transmitted to a data processing unit.

[0012] Furthermore, the process of performing distributed measurement of the temperature and pressure of the reactor wall is as follows:

[0013] Assign spatial coordinates to each sensor, determine the specific position of the sensor on the reactor wall based on the three-dimensional structure of the reactor, perform fitting of the corresponding temperature and pressure measurement values ​​to the data points, construct a cubic polynomial function between every two adjacent nodes, and determine the polynomial coefficients. At the nodes, the values ​​of adjacent piecewise polynomial functions are equal, the rates of change are continuous, and the second-order derivatives of adjacent piecewise polynomials at the nodes are equal. nodes, respectively , the corresponding temperature or pressure value is ;

[0014] Construct a polynomial function between every two adjacent nodes After finding the coefficients of all piecewise polynomials, for any position x on the reactor wall, first determine where x is in the interval , and substitute the corresponding piecewise polynomial The temperature or pressure value at any position is calculated; by calculating the position on the reactor wall, continuous and smooth temperature field or pressure field distribution data is obtained, and temperature field and pressure field distribution images are generated according to color mapping.

[0015] Furthermore, the process of converting the spectral data into the reaction data of the cross-linking agent intermediate is:

[0016] S201: Prepare cross-linker intermediate samples covering different reaction stages and component ratios, and measure the functional group conversion rate as a reference standard for subsequent model construction;

[0017] S202: Collecting spectral data of the above samples one by one;

[0018] S203: Using the partial least squares method, a correlation model between the spectral data and the reaction data is established; during the training process, the sample data is divided into a training set and a validation set, and the model parameters are continuously adjusted.

[0019] Furthermore, the process of establishing the correlation model between the spectral data and the reaction data is as follows:

[0020] The spectral data matrix is ​​represented by a matrix S with dimensions m × k, where m is the number of samples and k is the number of spectral wavelengths. The reaction data matrix is ​​represented by a matrix R with dimensions m × l, where l is the number of reaction data variables. The spectral data matrix S and the reaction data matrix R are standardized. An initial weight vector u is randomly selected with a length of cd corresponding to the number of spectral data variables. The weight vector u is normalized so that its modulus is 1.

[0021] Through iterative calculation, the score vector s1 and the load vector r1 are obtained; among them, s1 is the score of the first latent variable on the sample, and r1 reflects the degree of correlation between the spectral data variable and the first latent variable; based on the covariance of the score vector s1 and the spectral data matrix S, combined with the load vector r1, the weight update equation is constructed, and the new weight vector u is obtained by solving it. This process is repeated until the weight vector u converges; after obtaining the first latent variable, the residual calculation is performed.

[0022] Furthermore, the specific process of performing the residual calculation is as follows: after the residual calculation is performed, the above steps are repeated on the residual moments E1 and F1, m latent variables are taken, and the score matrix T and the load matrix F are obtained; the sample data are divided into z groups, and the model is trained with z-1 groups of data each time, and the remaining group of data is used to verify the model's prediction ability, and the prediction error is calculated, and this is repeated z times to obtain the residual.

[0023] Furthermore, the process of identifying the status data of the cross-linking agent intermediate product is:

[0024] Set the temperature threshold according to the normal temperature range in the cross-linking agent production process ; Temperature field image Perform threshold segmentation and separate the image with a temperature value greater than The pixels are marked as suspected overheated pixels; the features of the local overheated area are extracted and the pressure field image is calculated. Statistical characteristics of pressure data, using edge detection algorithm to detect edge areas with drastic pressure changes in pressure field images; through time series construction: collect temperature field images at different times in chronological order and pressure field images , construct time series data of temperature and pressure.

[0025] Furthermore, the process of training the ML model with real-time IoT data is as follows:

[0026] Based on the reaction mechanism of the cross-linker, a set of differential equations including the reaction rate equation and activation energy parameters was established. The historical data were used to modify the mechanism model parameters to form a semi-mechanism model. An ML model was established based on Fourier's law. The wall temperature data was used as the boundary condition. The reaction process inverted from the spectral data was coupled with the heat conduction model to calculate the impact of reaction heat release on the temperature field.

[0027] The cross-linking agent production control method based on the Internet of Things includes the following steps:

[0028] Step 1: Acquire spectral data inside the reactor, and acquire temperature data and pressure data at different locations on the reactor wall;

[0029] Step 2: Based on the meta-calibration model, the spectral data is converted into reaction data of the cross-linker intermediate product; based on the temperature and pressure data at different locations, image recognition and pattern analysis algorithms are used to identify the state data of the cross-linker intermediate product, including local overheating areas and abnormal pressure areas;

[0030] Step 3: Based on the reaction data and state data, a dynamic model of the production characteristics of the fusion cross-linker is established. Based on the dynamic model, the ML model is trained using historical data and real-time IoT data to automatically generate and execute the optimal control strategy under safety constraints.

[0031] The cross-linking agent production control system and method based on the Internet of Things provided by the present invention have the following beneficial effects:

[0032] (1) The present invention quantitatively tracks the changes in the intensity of characteristic peaks, records the absorbance of the characteristic peaks of the target reactants, and monitors the reaction progress of the cross-linking agent in real time, thereby replacing or significantly reducing offline laboratory testing and eliminating detection lag. This allows operators to grasp the reaction progress in a timely manner and provide an accurate basis for production decisions. The flexible sensor array realizes distributed monitoring of the temperature and pressure fields of the reactor, and can promptly detect potential risks such as local overheating and pressure anomalies that are difficult to detect with traditional single-point sensors. The system can also issue early warnings in a timely manner, effectively preventing the occurrence of safety accidents and improving production safety.

[0033] (2) The present invention performs distributed measurement through a flexible sensor array on the outer surface of the reactor wall to obtain temperature and pressure data at different positions, and generates continuous and smooth temperature field and pressure field distribution images through cubic polynomial function fitting; then combines image recognition and pattern analysis algorithms to accurately identify local overheating areas and abnormal pressure areas, and can timely discover potential risks that are difficult to detect with traditional single-point sensors, effectively prevent safety accidents, and improve production safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of the system of the present invention;

[0035] Figure 2 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0036] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0037] Example 1:

[0038] See also Figure 1 , Example 1 of the present application provides a cross-linking agent production control system based on the Internet of Things, the system comprising:

[0039] The data acquisition module acquires the spectral data inside the reactor, as well as the temperature and pressure data at different locations on the reactor wall;

[0040] Spectral data:

[0041] A near-infrared spectrometer is installed at a suitable location inside the reactor. The near-infrared spectrometer transmits near-infrared light through the materials in the reactor, receives the absorption and scattering signals of the materials to light of different wavelengths, and converts them into spectral data. Through the dynamic changes of the spectral characteristic peaks, a quantitative correlation is established between the spectral data and the functional group conversion rate, and the cross-linker reaction progress is monitored in real time. The near-infrared spectrometer collects spectral data at a frequency of multiple times per second to achieve continuous monitoring of the reaction progress. Its wavelength range covers 900-2500nm and can accurately capture the absorption spectral signals generated by the molecular vibration and rotation of the cross-linker intermediates during the reaction process. The spectrometer is equipped with a highly sensitive InGaAs detector and can achieve a data acquisition frequency of 5 times per second, ensuring timely acquisition of information on component changes in the reactor.

[0042] Crosslinker functional group conversion rate:

[0043] Infrared spectroscopy works by detecting the vibrational absorption of specific chemical bonds in molecules. In cross-linker synthesis, the disappearance of reactant functional groups or the appearance of product functional groups will directly lead to the intensity of their corresponding characteristic absorption peaks. The conversion rate is usually expressed as the percentage of reactant consumption or product formation.

[0044] Select appropriate characteristic peaks. The selected characteristic peaks should be clearly and uniquely attributed to the target reactant or target product. They should avoid overlap with peaks of other components, such as solvents, other reactants, and by-products. The characteristic peak intensity should be large enough to facilitate accurate measurement, but not too strong to cause signal saturation. The position and shape of the characteristic peak should be relatively stable under the reaction conditions, temperature, pressure, and concentration range, and be less susceptible to interference. The characteristic peak should preferably be located in a relatively "clean" area of ​​the spectrum with low background interference.

[0045] Examples of common functional groups in crosslinkers:

[0046] Monitoring the double bond consumption of acrylate crosslinkers, C=C stretching vibration peak (1630cm -1 ); As the reaction progresses, free radicals polymerize and double bonds open, and the intensity of this peak weakens;

[0047] At the same time, determine the zero point of the peak intensity and eliminate background interference, such as solvent absorption, light scattering, and instrument noise; select points with no absorption or stable absorption on both sides of the characteristic peak and connect them with a straight line or an appropriate curve to serve as the baseline of the peak;

[0048] It should be noted that an inaccurate baseline will directly lead to errors in peak intensity calculation, thereby affecting the accuracy of the conversion rate results;

[0049] By quantitatively tracking the changes in the intensity of the characteristic peaks, the conversion rate of the reaction can be calculated; the conversion rate calculation method based on the consumption of reactants:

[0050] Initial measurement: At the beginning of the reaction (time zero), the absorbance or peak area of ​​the characteristic peak of the target reactant, such as the absorption peak of a specific functional group, is recorded by the spectrometer;

[0051] Real-time monitoring: At a certain time point in the reaction, such as the tth minute, the absorbance or peak area of ​​the same characteristic peak is measured again;

[0052] Comparative analysis: Calculate the consumption ratio of reactants by comparing initial values ​​and real-time measurements; for example:

[0053] If the initial peak area of ​​the reactant is 100% and the real-time measurement is 60%, then the reactant has been consumed by 40%, corresponding to a conversion rate of 40%;

[0054] Calculation method of conversion rate based on product generation:

[0055] Product monitoring: At a certain time point in the reaction, such as the tth minute, the absorbance or peak area of ​​the characteristic peak of the target product, such as the absorption peak of the newly generated functional group, is recorded by spectrometer;

[0056] Determination of maximum value: By extending the reaction time, heating or adding excess reactants, the reaction is completely completed, and the maximum absorbance or peak area of ​​the characteristic peak of the product is measured, which is the theoretical maximum value. This value can also be obtained through standard sample experiments.

[0057] Comparative analysis: Compare the real-time product peak area with the maximum value; for example:

[0058] If the real-time measured product peak area is 60% and the maximum value is 100%, the product formation ratio is 60%, and the corresponding conversion rate is 60%;

[0059] Temperature data and pressure data:

[0060] The flexible sensor array is tightly attached to the outer surface of the reactor wall. The flexible sensor array consists of multiple micro temperature sensors and pressure sensors, which can perform distributed measurement of the temperature and pressure of the reactor wall. The sensors can obtain temperature and pressure data at different locations on the reactor wall in real time, forming a distribution image of the temperature and pressure fields of the reactor wall, thereby accurately capturing local overheating, pressure anomalies, etc., greatly improving the comprehensiveness and accuracy of monitoring compared to traditional single-point sensors.

[0061] Sensors convert the physical quantities of temperature and pressure into electrical signals. After amplification and filtering by signal conditioning circuits, they are output as digital signals and transmitted to the data processing unit. Each sensor is assigned unique spatial coordinates. Based on the three-dimensional structure of the reactor, the specific position of the sensor on the reactor wall is determined to accurately restore the temperature and pressure distribution. The 3σ principle or machine learning algorithms are used to identify and process abnormal data points. For example, if a temperature data point exceeds the normal fluctuation range, it can be replaced by the mean of the nearby normal data points or an estimated value based on the prediction model.

[0062] It should be noted that:

[0063] The signal transmission line of the sensor array uses a high-temperature resistant, flexible shielded cable, which is led out from the threading hole reserved in the reactor wall. During the threading process, the cable is treated with a protective sleeve to prevent cable wear and high-temperature burns. The cable is fixed to the reactor bracket with a cable tie or clamp to ensure neat wiring and prevent displacement due to equipment vibration, thereby avoiding pulling on the sensor array.

[0064] Since the sensors are discretely distributed on the reactor wall, interpolation calculation is required to obtain continuous temperature and pressure field distribution images;

[0065] Determine the installation position of the flexible sensor array on the reactor wall, perform fitting of the corresponding temperature and pressure measurement values, construct a cubic polynomial function between every two adjacent nodes, and determine the polynomial coefficients. The nodes are the installation positions of the flexible sensor array on the reactor wall:

[0066] At the nodes, the values ​​of adjacent piecewise polynomial functions are equal, ensuring the continuity of the function at the nodes and preventing jumps. The rate of change at the nodes is continuous, ensuring the smoothness of the curve and avoiding sharp corners. The second-order derivatives of adjacent piecewise polynomials at the nodes are equal, enhancing the smoothness of the function and ensuring the continuous curvature of the curve at the nodes.

[0067] Polynomial function is a mathematical expression used to describe the relationship between different positions on the reactor wall and the corresponding temperature or pressure values;

[0068] have nodes, respectively , the corresponding temperature or pressure value is For example, there are five sensors on the reactor wall, and their position coordinates and measured temperature values ​​constitute the nodes and corresponding data;

[0069] Construct a polynomial function between every two adjacent nodes After finding the coefficients of all piecewise polynomials, for any position x on the reactor wall (as long as x is within the interval covered by the node), first determine which interval x is located in , and then substitute the corresponding piecewise polynomial The temperature or pressure value at that location is calculated; by performing calculations on a large number of locations on the reactor wall, continuous and smooth temperature field or pressure field distribution data is obtained, and temperature field and pressure field distribution images are generated using a color mapping method; spline interpolation is used to reflect the changing trend of the data, resulting in a smoother temperature field and pressure field distribution; by determining the coefficients of the polynomial function, the temperature or pressure value at any location on the reactor wall is accurately calculated by substituting the polynomial function of the corresponding interval into the polynomial function, thereby obtaining continuous and smooth temperature field or pressure field distribution data for the entire reactor wall;

[0070] The continuous temperature and pressure data obtained through interpolation calculations are mapped to image pixel values. The temperature or pressure values ​​are associated with colors, for example, setting low temperature (low pressure) to blue, high temperature (high pressure) to red, and intermediate values ​​to different transition colors. Based on the spatial coordinate information of the reactor wall, the color value corresponding to the temperature or pressure at each position is filled into the corresponding pixel position, thereby generating temperature and pressure field distribution images.

[0071] The data processing and control module converts the acquired spectral data into reaction data of the crosslinker intermediate based on the meta-calibration model. Based on the temperature and pressure data at different locations, it uses image recognition and pattern analysis algorithms to identify the status data of the crosslinker intermediate, including local overheating areas, abnormal pressure areas, and intermediate product change trends.

[0072] Convert spectral data into reaction data of crosslinker intermediates:

[0073] S201: Prepare a large number of cross-linker intermediate product samples covering different reaction stages and component ratios, and measure the functional group conversion rate as a reference standard for subsequent model construction;

[0074] S202: Using a near-infrared spectrometer or other equipment, collect spectral data for each of the above samples, ensuring that collection conditions, such as wavelength range, resolution, and number of scans, are consistent to ensure data comparability;

[0075] S203: Using the partial least squares method, a correlation model between the spectral data and the reaction data is established; during the training process, the sample data is divided into a training set and a validation set according to a certain ratio, and the model parameters are continuously adjusted to ensure that the model has good prediction accuracy and generalization ability on the validation set;

[0076] Build a correlation model between spectral data and reaction data:

[0077] The spectral data matrix is ​​represented by a matrix S with dimensions m × k, where m is the number of samples and k is the number of spectral wavelengths. The reaction data matrix is ​​represented by a matrix R with dimensions m × l, where l is the number of reaction data variables. The spectral data matrix S and the reaction data matrix R are standardized. An initial weight vector u is randomly selected with a length of cd corresponding to the number of spectral data variables. The weight vector u is normalized so that its modulus is 1.

[0078] Through iterative calculation, the score vector s1 and the load vector r1 are obtained; where s1 is the score of the first latent variable on the sample, and r1 reflects the degree of correlation between the spectral data variable and the first latent variable; based on the covariance of the score vector s1 and the spectral data matrix S, combined with the load vector r1, a weight update equation is constructed and solved to obtain a new weight vector u. This process is repeated until the weight vector u converges; after obtaining the first latent variable, the residual calculation is performed;

[0079] Repeat the above steps on the residual moments E1 and F1, extracting the second, third, and so on latent variables until the stopping condition is met, that is, the cumulative explained variance reaches a certain proportion, usually 80%-95%; extract m latent variables, and obtain the score matrix T and the loading matrix F; use cross-validation and other methods to evaluate model performance, divide the sample data into z groups, and each time use z-1 groups of data to train the model, and use the remaining group of data to verify the model's predictive ability, calculate the prediction error, and repeat this process z times to obtain the residual error, and judge the accuracy and generalization ability of the model;

[0080] For new spectral data samples, they are processed according to the standardization method, and then their scores on the latent variables are calculated. The reaction data are predicted according to the established regression model, and the predicted standardized reaction data are restored to the actual reaction data.

[0081] Identify status data for cross-linker intermediates:

[0082] Temperature and pressure data integration:

[0083] The temperature and pressure data at different locations obtained from the flexible sensor array are integrated, and the temperature data matrix T and the pressure data matrix F are constructed according to the spatial coordinate relationship of the sensors on the reactor wall; the elements in the matrix correspond to the values ​​measured by the sensors at different locations, for example represents the temperature value measured by the sensor at the i-th row and j-th column position;

[0084] Perform median filtering on the temperature and pressure data. For each element in the temperature data matrix T, take its neighborhood, such as The median value of the element in the neighborhood is used to replace the element, thereby smoothing the data, eliminating isolated noise points, and removing noise interference; the temperature and pressure data are normalized to a specific range, such as [0,1], to facilitate subsequent processing and comparison;

[0085] Temperature field and pressure field image construction:

[0086] Convert the preprocessed temperature data matrix T and pressure data matrix F into image form; map the values ​​of the matrix elements to the grayscale value or color value of the image pixels. For example, the higher the temperature, the more red the pixel color; the greater the pressure, the more blue the color; and so on. Arrange the pixels in the order of the spatial coordinates of the sensor to generate a temperature field image. and pressure field images ;

[0087] Identification of local overheating areas:

[0088] Threshold segmentation: Set the temperature threshold according to the normal temperature range in the cross-linking agent production process ; Temperature field image Perform threshold segmentation and separate the image with a temperature value greater than Pixels with the same temperature as the one in the suspected overheating area are marked as pixels in the suspected overheating area; for example, if the normal temperature range is , you can Set as ,The pixel area above this temperature may be an overheated area;

[0089] Morphological processing:

[0090] Morphological operations, such as dilation and erosion, are performed on the labeled image. The dilation operation connects adjacent pixels in suspected overheated areas to form a more complete region, while the erosion operation removes isolated small noise areas. Through these operations, the final local overheated area is determined. Regional feature extraction is performed to extract the characteristics of the local overheated area, such as area, shape, and location. These features can be used to further analyze the severity of the overheated area and its impact on the crosslinker reaction.

[0091] Identify abnormal pressure areas and calculate pressure field images Statistical characteristics of medium pressure data, such as mean and standard deviation According to production experience or process requirements, determine the judgment standard of pressure abnormality, for example, the area where the pressure value deviates from the mean by more than k times the standard deviation, k is usually (2-3) is the pressure abnormal area;

[0092] Through edge detection, using edge detection algorithms such as the Canny operator and the Sobel operator, we can detect edge areas in the pressure field image where pressure changes dramatically. These areas may be the boundaries of pressure anomalies. Combined with statistical analysis results, we can determine the scope of the pressure anomaly area.

[0093] Analysis of intermediate product change trends, constructed through time series: collecting temperature field images at different times in chronological order and pressure field images , constructing time series data of temperature and pressure; extracting features related to changes in intermediate products from the time series images. For example, calculating the rate of change of the overheated area in the temperature field image over time, and the movement direction and speed of the pressure anomaly area in the pressure field image. Linear regression is used to analyze the relationship between the rate of change of the overheated area and the conversion rate of the intermediate products. The extracted features are then used to build models to predict the changing trends of the intermediate products and their conversion trends over a period of time.

[0094] Identified local overheating areas and abnormal pressure areas are presented in the form of visual graphics, such as markings on the reactor schematic. At the same time, the trend of intermediate product changes is output in the form of charts (such as a line chart showing the change in conversion rate over time) or text, providing a basis for monitoring and adjustment of the cross-linker production process;

[0095] The execution decision module builds a dynamic model of the cross-linker production characteristics based on reaction data, temperature data, and pressure data. It uses historical data and real-time IoT data to train the ML model and automatically generates and executes the optimal control strategy under safety constraints.

[0096] Develop a dynamic model of the production characteristics of fusion crosslinkers:

[0097] The spectral data is aligned with the temperature and pressure data using timestamps, and the reactor wall temperature and pressure data are mapped to a three-dimensional spatial coordinate system. A spatial association is established with the spectral data inside the reactor. The wall temperature is mapped to the internal temperature field through a heat conduction model to obtain spectral features, as well as temperature and pressure features. Cross-features are then performed to construct interactive features of the spectrum, temperature, and pressure, namely the product of absorbance at a specific wavelength and local temperature, to capture the physical and chemical coupling effect. Based on the cross-linker reaction mechanism, a set of differential equations including the reaction rate equation and activation energy parameters is established. Utilizing historical data, such as conversion rate and molecular weight distribution detected offline, a semi-mechanistic model is formed by modifying the mechanistic model parameters. An ML model is established based on Fourier's law, using the wall temperature data as a boundary condition. The reaction process inverted from the spectral data is coupled with the heat conduction model to calculate the effect of reaction heat release on the temperature field.

[0098] By obtaining the reactant concentration, temperature field distribution, pressure field distribution and reaction conversion rate, a state equation is established. Through the time derivative of the state variable, the state vector, control input and disturbance are used as independent variables to include various state information related to crosslinker production in the reactor, such as reactant concentration, temperature field distribution, pressure field distribution, reaction conversion rate, etc. The control input is a manually adjustable operating variable, such as the feed flow rate and temperature setting value, and the disturbance is an external uncertain factor that will affect the production process. The mechanism model structure is combined with the data-driven method, and the LSTM network is used to learn the unmodeled dynamics in the mechanism model, such as side reactions, and to modify uncertain parameters such as the thermal conductivity coefficient. The model parameters are optimized using gradient descent or genetic algorithms by minimizing the error between the model prediction value and the actual value, such as the deviation between the spectral inversion concentration and the temperature field.

[0099] Real-time IoT data training ML models:

[0100] The trained ML model is combined with safety constraints and the optimization objective function, and an optimization algorithm is used. Taking model predictive control as an example, in each control cycle, based on the current production status, provided by real-time IoT data, the ML model is used to predict the changing trend of the production process in the future. Under the premise of meeting the safety constraints, the optimization problem is solved to obtain the control input that optimizes the objective function, such as feed flow rate, temperature set point, stirring speed, etc., thus generating the optimal control strategy.

[0101] Automatically generate and execute optimal control strategies under safety constraints:

[0102] The generated optimal control strategy is sent to the actuators of the crosslinker production equipment, such as the feed pump, heating / cooling device, and agitator, to adjust the operating parameters of the production process in real time;

[0103] Continuously monitor real-time IoT data during the production process to identify deviations between the actual production status and the expected status. This feedback is promptly transmitted to the execution decision module to adjust and optimize the control strategy. Based on this feedback, determine whether the current control strategy is achieving the expected results. If the actual production status deviates from the target, analyze the cause of the deviation to determine whether it is due to inaccurate model predictions or new interference factors in the production process. Based on the analysis results, update the ML model, such as retraining the model with new real-time data or regenerating the optimal control strategy. This allows for dynamic adjustment and iterative optimization of the control strategy, ensuring that the crosslinker production process always operates safely and efficiently.

[0104] Example 2

[0105] See also Figure 2 Based on Example 1, Example 2 of the present application further provides a cross-linking agent production control method based on the Internet of Things, including the following specific steps:

[0106] Step 1: Acquire spectral data inside the reactor, and acquire temperature data and pressure data at different locations on the reactor wall;

[0107] Step 2: Based on the meta-calibration model, the spectral data is converted into reaction data of the cross-linker intermediate product; based on the temperature and pressure data at different locations, image recognition and pattern analysis algorithms are used to identify the state data of the cross-linker intermediate product, including local overheating areas and abnormal pressure areas;

[0108] Step 3: Based on the reaction data and state data, a dynamic model of the production characteristics of the fusion cross-linker is established. Based on the dynamic model, the ML model is trained using historical data and real-time IoT data to automatically generate and execute the optimal control strategy under safety constraints.

[0109] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0110] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0111] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. The cross-linking agent production control system based on the Internet of Things is characterized by: The system includes: The data acquisition module acquires the spectral data inside the reactor, as well as the temperature and pressure data at different locations on the reactor wall; The spectral data is obtained by quantitatively tracking the change in the intensity of the characteristic peak, recording the absorbance of the characteristic peak of the target reactant, and measuring the absorbance of the same characteristic peak again at the tth minute of the reaction; by comparing the initial value and the real-time measurement value, calculating the consumption ratio of the reactant, until the reaction is completely completed, and measuring the maximum absorbance of the characteristic peak of the product, i.e., the spectral data, to monitor the progress of the crosslinker reaction in real time; The data processing and control module converts spectral data into reaction data of the crosslinker intermediate based on the meta-calibration model. Based on the temperature and pressure data at different locations, it uses image recognition and pattern analysis algorithms to identify the status data of the crosslinker intermediate, including local overheating areas and abnormal pressure areas. The process of converting spectral data into reaction data of a cross-linking agent intermediate is as follows: S201: preparing cross-linking agent intermediate samples covering different reaction stages and component ratios, and measuring functional group conversion rates as reference standards for subsequent model construction; S202: collecting spectral data for the above samples one by one; S203: using partial least squares method to establish a correlation model between spectral data and reaction data; during the training process, the sample data is divided into a training set and a validation set, and the model parameters are continuously adjusted; The execution decision module establishes a dynamic model of the production characteristics of the fusion cross-linker based on reaction data and status data. Based on the dynamic model, the ML model is trained using historical data and real-time IoT data to automatically generate and execute the optimal control strategy under safety constraints.

2. The cross-linking agent production control system based on the Internet of Things according to claim 1, characterized in that: The process of obtaining temperature data and pressure data at different positions on the reactor wall is as follows: The temperature and pressure of the reactor wall are measured in a distributed manner through a flexible sensor array on the outer surface of the reactor wall; the measurements are output in the form of digital signals and transmitted to a data processing unit.

3. The cross-linking agent production control system based on the Internet of Things according to claim 2, characterized in that: The process of performing distributed measurement of the temperature and pressure of the reactor wall is as follows: Assign spatial coordinates to each sensor, determine the specific position of the sensor on the reactor wall based on the three-dimensional structure of the reactor, perform fitting of the corresponding temperature and pressure measurement values ​​to the data points, construct a cubic polynomial function between every two adjacent nodes, and determine the polynomial coefficients. At the nodes, the values ​​of adjacent piecewise polynomial functions are equal, the rates of change are continuous, and the second-order derivatives of adjacent piecewise polynomials at the nodes are equal. nodes, respectively , the corresponding temperature or pressure value is ; Construct a polynomial function between every two adjacent nodes After finding the coefficients of all piecewise polynomials, for any position x on the reactor wall, first determine the interval where x is located and substitute it into the corresponding piecewise polynomial The temperature or pressure value at any position is calculated; by calculating the position on the reactor wall, continuous and smooth temperature field or pressure field distribution data is obtained, and temperature field and pressure field distribution images are generated according to color mapping.

4. The cross-linking agent production control system based on the Internet of Things according to claim 3, characterized in that: The process of establishing the correlation model between the spectral data and the reaction data is as follows: The spectral data matrix is ​​represented by a matrix S with dimensions m × k, where m is the number of samples and k is the number of spectral wavelengths. The reaction data matrix is ​​represented by a matrix R with dimensions m × l, where l is the number of reaction data variables. The spectral data matrix S and the reaction data matrix R are standardized. An initial weight vector u is randomly selected with a length of cd corresponding to the number of spectral data variables. The weight vector u is normalized so that its modulus is 1. Obtain the score vector through iterative calculation and the load vector ;in, is the score of the first latent variable on the sample, Reflects the degree of correlation between the spectral data variable and the first latent variable; based on the score vector The covariance with the spectral data matrix S, combined with the loading vector , construct the weight update equation, solve it to get the new weight vector u, repeat this process until the weight vector u converges; after obtaining the first latent variable, perform residual calculation.

5. The cross-linking agent production control system based on the Internet of Things according to claim 4, characterized in that: The specific process of performing residual calculation is: After calculating the residuals, repeat the above steps on the obtained residual moments E1 and F1, take m latent variables, and obtain the score matrix T and the loading matrix F; divide the sample data into z groups, use z-1 groups of data to train the model each time, use the remaining group of data to verify the model's prediction ability, calculate the prediction error, and repeat this z times to obtain the residuals.

6. The cross-linking agent production control system based on the Internet of Things according to claim 1, characterized in that: The process of identifying the status data of the cross-linking agent intermediate product is: Set the temperature threshold according to the normal temperature range in the cross-linking agent production process ; Temperature field image Perform threshold segmentation and separate the image with a temperature value greater than The pixels are marked as suspected overheating pixels; Extract the features of local overheating areas and calculate the pressure field image Statistical characteristics of pressure data, using edge detection algorithm to detect edge areas with drastic pressure changes in pressure field images; through time series construction: collect temperature field images at different times in chronological order and pressure field images , construct time series data of temperature and pressure.

7. The cross-linking agent production control system based on the Internet of Things according to claim 1, characterized in that: The process of training the ML model with real-time IoT data is as follows: Based on the reaction mechanism of the cross-linker, a set of differential equations including the reaction rate equation and activation energy parameters was established. The historical data were used to modify the mechanism model parameters to form a semi-mechanism model. An ML model was established based on Fourier's law. The wall temperature data was used as the boundary condition. The reaction process inverted from the spectral data was coupled with the heat conduction model to calculate the impact of reaction heat release on the temperature field.

8. A cross-linking agent production control method based on the Internet of Things, characterized in that: The steps include: Step 1: Acquire spectral data inside the reactor, and acquire temperature data and pressure data at different locations on the reactor wall; The spectral data is obtained by quantitatively tracking the change in the intensity of the characteristic peak, recording the absorbance of the characteristic peak of the target reactant, and measuring the absorbance of the same characteristic peak again at the tth minute of the reaction; by comparing the initial value and the real-time measurement value, calculating the consumption ratio of the reactant, until the reaction is completely completed, and measuring the maximum absorbance of the characteristic peak of the product, i.e., the spectral data, to monitor the progress of the crosslinker reaction in real time; Step 2: Based on the meta-calibration model, the spectral data is converted into reaction data of the cross-linker intermediate product; based on the temperature and pressure data at different locations, image recognition and pattern analysis algorithms are used to identify the state data of the cross-linker intermediate product, including local overheating areas and abnormal pressure areas; The process of converting spectral data into reaction data of a cross-linking agent intermediate is as follows: S201: preparing cross-linking agent intermediate samples covering different reaction stages and component ratios, and measuring functional group conversion rates as reference standards for subsequent model construction; S202: collecting spectral data for the above samples one by one; S203: using partial least squares method to establish a correlation model between spectral data and reaction data; during the training process, the sample data is divided into a training set and a validation set, and the model parameters are continuously adjusted; Step 3: Based on the reaction data and state data, a dynamic model of the production characteristics of the fusion cross-linker is established. Based on the dynamic model, the ML model is trained using historical data and real-time IoT data to automatically generate and execute the optimal control strategy under safety constraints.

Images